I have M*N dataset where M=Samples and N=features. i want to select features due to which SVM classification accuracy is high by using Particle swarm optimization .
i am not getting
1) where to assign features of dataset in pso implementation?[I mean how to load dataset for PSO]
2) swarm size[is it equal to M = no. of sample ?]
3) Where to assign M in PSO algorithm?
I have googled a lot and stuck over here in my research work . thanks a lot for help .
Hello Suman,
at first, I'd rather use a genetic algorithm for this kind of task...it would be easier to have an individual as a binary string representing the acceptance of feature into your subset (110110 - will mean that 1,2,4 and 5ht features are selected) - in case of PSO, I'm not sure if there is some binary representation. There are some discrete but it will not help you much.
so I recommend you to use GA + SVM - and the number of candidates in GA should not be higher than 2* number of features...
good luck,
regards,
tomas
i think you may need to follow the following steps;
1. select N random initialisation samples from from your data set, these will be your swarm particles
2. implement the PSO algorithm
3. at the end of your PSO, analyse the behaviour of each swarm to determine the feature (or features) that best represents the swarm.
@AAbass Olaode : thank you sir , did you mean particle = M*N matrix [Trainging dataset ] or it is [1*N] matrix (1 samples at a times with all features)
note M is the number of samples in your set, while N is the number of features. By picking N random samples (where N is the number of features) at the beginning of the PSO implementation, you are creating the opportunity for the samples to group themselves based on the features they have in common. At the end of the PSO the features common to each group will be revealed through their values.
you can use the final swarm particles as your training set in the implementation of SVM .
One option could be the PySwarms toolkit:
http://pyswarms.readthedocs.io/en/latest/index.html
In particular, to feature selection with PSO:
http://pyswarms.readthedocs.io/en/latest/examples/feature_subset_selection.html
Best regards
@ Abass Olaode : Thanks a lot
rows=no of samples
X_train = dataset (Sample * features) matrix
Is this is correct way to group particles as you said?
Kindly have a look at the following URL for feature selection in Python using PSO (Python code):
https://ljvmiranda921.github.io/projects/2017/08/11/pyswarms/
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages
- Python